--- title: Multiplicative Trace Preservers url: https://www.emergentmind.com/topics/multiplicative-trace-preservers type: topic --- # Multiplicative Trace Preservers Multiplicative trace preservers are maps, families of maps, or natural transformations that preserve a trace-compatible multiplicative structure. The term is not uniform across the literature. In higher algebra it denotes multiplicative natural transformations out of algebraic \(K\)-theory, notably the topological Dennis trace and the cyclotomic trace. In matrix preserver theory it denotes maps preserving expressions such as \(\operatorname{tr}(\phi(A)\psi(B)^k)=\operatorname{tr}(AB^k)\), \(\operatorname{tr}(\phi_1(A_1)\cdots \phi_m(A_m))=\operatorname{tr}(A_1\cdots A_m)\), or the trace of a Kronecker sum. In quantum information it is tied to trace-preserving completely positive maps and their multiplicative domains. In the theory of linear groups it refers to homomorphisms preserving the ordinary matrix trace on products in \(SL(2,\mathbb C)\). Across these settings, the common theme is rigidity: trace preservation together with multiplicative compatibility typically forces strong canonical forms or uniqueness statements [1103.3923], [2103.12552], [1701.06205], [1608.08212]. ## 1. Terminological scope and basic formulations The most general pattern is that one asks for maps compatible with multiplication while leaving a trace-type invariant unchanged. In matrix settings, a standard form is \[ \operatorname{tr}(\phi_1(A_1)\cdots \phi_m(A_m))=\operatorname{tr}(A_1\cdots A_m), \] or, with powers, \[ \operatorname{tr}(\phi(A)\psi(B)^k)=\operatorname{tr}(AB^k), \] for prescribed classes of matrices. In tensor-structured problems one instead preserves \[ \operatorname{tr}(A\oplus B)=n\,\operatorname{tr}(A)+m\,\operatorname{tr}(B), \] where \(A\oplus B=A\otimes I_n+I_m\otimes B\). In higher algebra the multiplicativity condition is formulated through lax symmetric monoidal functors and multiplicative natural transformations. There one requires compatibility with monoidal structure diagrams, so that for a ring spectrum \(R\) the induced map on \(K(R)\) is a map of ring spectra [2201.09513], [1901.01720], [1103.3923]. These formulations differ substantially in ambient category, but they share two recurrent mechanisms. First, trace preservation is usually paired with a nondegenerate bilinear or monoidal structure, which converts the trace identity into linearity, injectivity, or uniqueness. Second, multiplicativity propagates local information to global structure: preserving a trace on selected products often forces similarity, congruence, permutation-conjugation, or an automorphic action on a distinguished subalgebra [2103.12552], [1701.06205]. A useful way to organize the subject is by ambient setting. | Setting | Preserved quantity | Typical outcome | |---|---|---| | Algebraic \(K\)-theory | Multiplicative natural transformations to \(THH\) or \(TC\) | Uniqueness and contractibility [1103.3923] | | Matrix preserver theory | \(\operatorname{tr}(AB^k)\), \(\operatorname{tr}(A_1\cdots A_m)\), or \(\operatorname{tr}(A\oplus B)\) | Similarity, congruence, or permutation forms [2201.09513], [2103.12552], [1901.01720] | | Quantum channels | Multiplicativity on a subalgebra for TP/CP maps | Characterization by multiplicative domain [1701.06205] | | \(SL(2,\mathbb C)\) groups | \(\operatorname{tr}(h(g))=\operatorname{tr}(g)\) | Innerness under fixed-point hypotheses [1608.08212] | This suggests that “multiplicative trace preserver” is best understood as a family resemblance term rather than a single formal definition. A plausible implication is that the literature uses the phrase to emphasize rigidity phenomena generated by the interaction of trace identities with multiplicative structure. ## 2. Multiplicative trace preservers in algebraic \(K\)-theory In the setting of noncommutative motives, multiplicative trace preservers are canonical multiplicative natural transformations out of algebraic \(K\)-theory landing in topological Hochschild homology \(THH\) or topological cyclic homology \(TC\). The ambient category is \(Cat^{\mathrm{perf}}\), the \(\infty\)-category of small idempotent-complete stable \(\infty\)-categories and exact functors. The connective and nonconnective algebraic \(K\)-theory functors are corepresentable in the motive categories \(\mathcal M_{\mathrm{add}}\) and \(\mathcal M_{\mathrm{loc}}\): \[ \operatorname{Map}(U_{\mathrm{add}}(S^\omega),U_{\mathrm{add}}(A))\simeq K(A),\qquad \operatorname{Map}(U_{\mathrm{loc}}(S^\omega),U_{\mathrm{loc}}(A))\simeq IK(A). \] The paper proves that \(K\) is the tensor unit in the symmetric monoidal \(\infty\)-category of additive invariants, and this tensor-unit role drives the trace uniqueness statements [1103.3923]. The central theorems are explicit. The space of multiplicative natural transformations from \(K\) to \(THH\) is contractible, \[ \operatorname{Map}^{\otimes}(K,THH)\simeq *, \] and the unique point is the multiplicative topological Dennis trace. Likewise, for each \(n\), \[ \operatorname{Map}^{\otimes}(K,TC^n)\simeq *, \] and the compatible lift through the tower yields the multiplicative cyclotomic trace \(tr_{\mathrm{cyc}}:K\Rightarrow TC\). The paper also proves that the space of multiplicative structures on \(K\) itself is contractible: \(\operatorname{Alg}_{E_n}(K)\simeq *\). Hence there is no ambiguity in the multiplicative refinement of algebraic \(K\)-theory [1103.3923]. The structural input is equally important. \(THH\) is realized as a lax symmetric monoidal localizing invariant, and each \(TC^n\) is multiplicative as well. The multiplicative Morita equivalence \[ (Cat^{\mathrm{perf}})^\otimes \simeq (N(Cats_{\mathrm{flat}})[W^{-1}])^\otimes \] identifies small idempotent-complete stable \(\infty\)-categories with the Morita localization of flat spectral categories in a symmetric monoidal way. Together with Glasman’s Day-convolution equivalence, this permits multiplicative natural transformations to be interpreted as morphisms of \(E_\infty\)-algebras in functor categories [1103.3923]. These results place the Dennis and cyclotomic traces in an unusually rigid position. Any multiplicative trace from algebraic \(K\)-theory to \(THH\) or \(TC\) must coincide with the standard one. The paper further shows that \(K(-)\) is lax symmetric monoidal, with structural maps \[ K(\mathcal A)\wedge K(\mathcal B)\to K(\mathcal A\otimes \mathcal B), \] and therefore if \(R\) is an \(E_n\)-ring spectrum, then \(K(R)\) is an \(E_{n-1}\)-ring spectrum. This “one-step down” multiplicative behavior is a direct consequence of the monoidal formalism [1103.3923]. ## 3. Matrix trace preservers on products, powers, and determinant-linked functionals A large matrix-preserver literature studies multiplicative trace preservers as maps on matrix spaces for which the trace of products, power-products, or determinant-derived expressions remains unchanged. One foundational form is the multiplicative trace functional on positive definite matrices, \[ f(A,B)=\operatorname{tr}(AB^{-1}),\qquad A,B\in P_n. \] If \(\phi:P_n\to P_n\) satisfies \[ \det(A+B)=\det(\phi(A)+\phi(B))\quad\text{for all }A,B\in P_n, \] then \[ \operatorname{tr}(AB^{-1})=\operatorname{tr}(\phi(A)\phi(B)^{-1}) \] for all \(A,B\in P_n\). The paper proves that this condition is equivalent to a congruence or transpose-congruence form: \[ \phi(A)=M^*AM\quad\text{or}\quad \phi(A)=M^*A^tM, \] with \(M\) invertible and \(\det(M^*M)=1\). Parallel classifications are obtained for symmetric matrices, general matrices, upper triangular matrices, and diagonal matrices, with the corresponding double-sided similarity, orthogonal congruence, or permutation-and-scaling forms [1603.03869]. A broader local-to-global classification appears for trace-of-power-product preservers. For matrix spaces \(V\) and fixed \(k\in\mathbb Z\setminus\{0,1\}\), one studies linear maps \(\phi,\psi:V\to V\) satisfying \[ \operatorname{tr}(\phi(A)\psi(B)^k)=\operatorname{tr}(AB^k). \] The paper first classifies local \(k\)-power preservers \(\psi\), namely maps with \(\psi(A^k)=\psi(A)^k\) on an open neighborhood of \(I_n\), and then deduces the form of \(\phi\) by trace-adjoint arguments. On \(M_n(F)\), for instance, such pairs have the form \[ \phi(A)=c^{-k}PAP^{-1},\qquad \psi(B)=cPBP^{-1}, \] or the transpose variants, when \(k\neq -1,0,1\). On Hermitian, symmetric, positive definite, diagonal, and upper triangular classes one obtains the corresponding unitary, orthogonal, congruence, permutation, or triangular similarity forms, with parity restrictions and positivity constraints where appropriate [2201.09513]. For multiplicative trace preservers involving \(m\) factors, \[ \operatorname{tr}(\phi_1(A_1)\cdots \phi_m(A_m))=\operatorname{tr}(A_1\cdots A_m), \] there is a marked rigidity jump when \(m\ge 3\). On \(M_n(F)\), the classification is \[ \phi_i(A)=N_i A N_{i+1}^{-1},\qquad N_{m+1}=N_1. \] On \(H_n\) and \(P_n\), odd \(m\) forces a common unitary congruence with real scalars \(c_i\) satisfying \(\prod_i c_i=1\), while even \(m\) yields alternating forms using a single invertible \(M\): \[ \phi_i(A)=c_iM^*AM\ \text{for odd }i,\qquad \phi_i(A)=c_iM^{-1}AM^{-*}\ \text{for even }i. \] For symmetric matrices the same pattern holds with transpose in place of adjoint, and for diagonal matrices one gets \[ \phi_i(A)=C_iP^tAP, \] with a permutation matrix \(P\) and diagonal invertible \(C_i\) satisfying \(C_1\cdots C_m=I\) [2103.12552]. The common methodological backbone is the nondegeneracy of the bilinear form \(\langle A,B\rangle=\operatorname{tr}(AB)\). In the two-map case this nondegeneracy forces linearity, injectivity, and dimension equalities; in the \(m\)-map case it combines with cyclicity of trace and algebraic telescoping identities to impose a global implementing matrix or permutation. This suggests that the principal source of rigidity is not merely multiplicativity, but multiplicativity filtered through a nondegenerate trace pairing [2103.12552], [2201.09513], [1603.03869]. ## 4. Tensor-structured preservers: Kronecker sums, partial traces, and partial determinants A distinct line of work studies preservers of the trace of the Kronecker sum \[ A\oplus B=A\otimes I_n+I_m\otimes B, \] for \(A\in M_m\), \(B\in M_n\). Its trace satisfies \[ \operatorname{tr}(A\oplus B)=n\,\operatorname{tr}(A)+m\,\operatorname{tr}(B). \] The classification is expressed in terms of partial traces. For \(X\in M_{mn}\), viewed as an \(m\times m\) block matrix with blocks in \(M_n\), the partial traces are \[ \operatorname{tr}_2(X)=\sum_{i,j=1}^m \operatorname{tr}(X_{ij})E_{ij}^{(m)},\qquad \operatorname{tr}_1(X)=\sum_{j=1}^m X_{jj}. \] If \(\phi(M)=PM\) is left multiplication, then \[ \operatorname{tr}(\phi(A\oplus B))=\operatorname{tr}(A\oplus B)\ \text{for all }A,B \] if and only if \[ \operatorname{tr}_1(P)=mI_n,\qquad \operatorname{tr}_2(P)=nI_m. \] For a general linear map \(\phi\), the same preserver property is equivalent to the corresponding partial-trace conditions on the index-swapped map \(\phi'\) evaluated at the identity [1901.01720]. The paper introduces RT-symmetry as the relevant tensor-order symmetry. If \[ \phi(X)=\sum_{j,k,u,v}\alpha_{jk;uv}\,E_{jk}^{(n)}XE_{uv}^{(n)}, \] then the index-swapped map \(\phi'\) interchanges the tensor indices. The map is RT-symmetric when \(\phi=\phi'\) and skew RT-symmetric when \(\phi=-\phi'\). These symmetries sharpen the preserver characterization by reducing it to explicit sign conditions on \(\operatorname{tr}_1(\phi(I_{mn}))\) and \(\operatorname{tr}_2(\phi(I_{mn}))\) [1901.01720]. The multiplicative aspect enters through the exponential–Kronecker identity \[ \det(e^A\otimes e^B)=e^{\operatorname{tr}(A\oplus B)}. \] Thus preserving \(\operatorname{tr}(A\oplus B)\) induces determinant-preserving behavior on Kronecker products. Partial determinants \(\operatorname{Det}_1\) and \(\operatorname{Det}_2\) then provide factorwise multiplicative constraints mirroring the partial-trace classification. In this sense, the additive invariant \(\operatorname{tr}(A\oplus B)\) exponentiates into a multiplicative invariant on tensor products [1901.01720]. This tensor-structured theory differs from the product-preserver setting because the underlying decomposition is bipartite. Partial trace and partial determinant replace the ordinary trace pairing as the decisive invariants. A plausible implication is that tensor-factor localization, rather than full-space linearity alone, is what makes the classification possible. ## 5. Stochastic matrices and reduction to the doubly stochastic component For stochastic matrices, multiplicative trace preservers are families \(\phi_i:\mathcal S\to\mathcal S\) such that \[ \operatorname{tr}(\phi_1(A_1)\cdots \phi_m(A_m)) = \operatorname{tr}(A_1\cdots A_m), \] or the analogous spectrum-preserving identity, where \(\mathcal S\) is one of \(DS_n\), \(RS_n\), \(CS_n\), or their linear spans. The decisive structural fact is that every matrix admits a canonical decomposition \[ A=A_{DS}+A_R+A_C, \] with \(A_{DS}\in\langle DS_n\rangle\), \(A_R\in R_n\), and \(A_C\in C_n\). For products over \(\langle RS_n\rangle\) or \(\langle CS_n\rangle\), the doubly stochastic components multiply according to \[ (A_1\cdots A_m)_{DS}=(A_1)_{DS}\cdots(A_m)_{DS}, \] and therefore \[ \operatorname{spec}(A_1\cdots A_m)=\operatorname{spec}\!\bigl((A_1)_{DS}\cdots(A_m)_{DS}\bigr),\qquad \operatorname{tr}(A_1\cdots A_m)=\operatorname{tr}\!\bigl((A_1)_{DS}\cdots(A_m)_{DS}\bigr). \] The paper states this as the central insight: the \(DS\)-component carries the spectral and trace information [2509.22743]. On \(DS_n\), the classification is explicit. For \(m=2\), multiplicative trace preservation and multiplicative spectrum preservation are equivalent, and one has either \[ \phi_1(A)=PAQ^t,\qquad \phi_2(A)=QAP^t, \] or the transpose version, with \(P,Q\in P_n\). For \(m\ge 3\), the forms are cyclic: \[ \phi_i(A)=P_iAP_{i+1}^t,\qquad P_{m+1}=P_1. \] On the span \(\langle DS_n\rangle\), one obtains inner or transpose-inner conjugations by invertibles in the algebra. On \(\langle RS_n\rangle\) and \(\langle CS_n\rangle\), the \(DS\)-component has the same form as on \(\langle DS_n\rangle\), while the non-\(DS\) parts are arbitrary tails \(\gamma_i\) landing in \(R_n\) or \(C_n\), subject only to the requirement that the image remain in the ambient space [2509.22743]. A particularly strong conclusion is that when \(m\ge 3\), multiplicative trace preservers always coincide with multiplicative spectrum preservers across all stochastic sets and spans considered. For \(m=1\) and \(m=2\), extra assumptions are needed in some span cases, but from three factors onward the two notions agree. This is a stochastic analogue of the rigidity phenomenon seen in general matrix products: additional multiplicative slots force the trace constraint to encode the full automorphic structure [2509.22743]. ## 6. Quantum channels, multiplicative domains, and trace-preserving complete positivity In quantum information theory, the ambient objects are quantum channels \(\Phi:M_n\to M_n\), that is, completely positive trace-preserving maps. The relevant multiplicative notion is not usually preservation of the trace of arbitrary products by a family of maps, but preservation of matrix multiplication on a distinguished subalgebra. Passing to the adjoint \(\Psi=\Phi^\dagger\), which is unital when \(\Phi\) is trace-preserving, one defines the multiplicative domain \[ MD(\Psi)=\{X\in M_n:\Psi(XY)=\Psi(X)\Psi(Y)\ \text{and}\ \Psi(YX)=\Psi(Y)\Psi(X)\ \forall Y\in M_n\}. \] An equivalent characterization is \[ MD(\Psi)=\{X:\Psi(X^\dagger X)=\Psi(X)^\dagger\Psi(X),\ \Psi(XX^\dagger)=\Psi(X)\Psi(X)^\dagger\}. \] Thus multiplicative behavior for trace-preserving channels is encoded by the multiplicative domain of the unital adjoint [1701.06205]. The paper gives several equivalent descriptions. For a unital channel \(E\), \[ MD(E)=\mathcal F_{E^*\circ E}, \] the fixed-point algebra of \(E^*\circ E\). If \(E(X)=\sum_j a_jXa_j^\dagger\), then \[ X\in MD(E)\iff Xa_i^\dagger a_j=a_i^\dagger a_jX\quad\text{for all }i,j. \] This yields a Kraus-level criterion for nontrivial multiplicativity. The eventual multiplicative domain \[ \mathcal M_\infty(E)=\bigcap_{k\ge 1}MD(E^k) \] is a stabilizing \(C^*\)-subalgebra on which \(E\) acts as a bijective \(*\)-homomorphism, while \(E\) is strictly contractive on the Hilbert–Schmidt orthogonal complement. The restriction \(E|_{\mathcal M_\infty(E)}\) has spectrum equal to the peripheral spectrum of \(E\), and \(\mathcal M_\infty(E)\) is the algebra generated by peripheral eigenoperators [1701.06205]. Global multiplicativity is extremely rigid. A trace-preserving CP map is multiplicative on all of \(M_n\) if and only if it is a \(*\)-automorphism, hence unitary conjugation \[ \Phi_U(X)=UXU^\dagger. \] More generally, a channel is multiplicative precisely on the subalgebra determined by its multiplicative domain. The multiplicative index \[ m(\Psi)=\min\{m\ge 0:MD(\Psi^m)=MD(\Psi^{m+1})\} \] measures stabilization time, with \(m=1\) for normal or diagonalizable channels such as Pauli channels, while explicit examples on \(M_3\) show larger values. The paper also proves that channels with trivial multiplicative domain are dense in completely bounded norm among unital CP maps [1701.06205]. Relative to matrix preserver theory, the quantum-channel literature replaces exact preservation of \(\operatorname{tr}(A_1\cdots A_m)\) by exact multiplicativity on a subalgebra detected through trace-duality. The common principle remains the same: multiplicativity is governed by a rigid invariant algebra, and trace preservation supplies the correct adjoint formalism for finding it. ## 7. Trace-preserving homomorphisms on \(SL(2,\mathbb C)\) and representation-theoretic rigidity For subgroups \(G\le SL(2,\mathbb C)\), a trace-preserving homomorphism is a group homomorphism \(h:G\to I\) such that \[ \operatorname{tr}(h(g))=\operatorname{tr}(g)\qquad \text{for all } g\in G. \] Because trace on \(SL(2,\mathbb C)\) carries strong conjugacy information, such homomorphisms behave like multiplicative trace preservers in a group-theoretic sense. The key identities are \[ \operatorname{tr}(XY)=\operatorname{tr}(YX),\qquad \operatorname{tr}(X^{-1})=\operatorname{tr}(X),\qquad \operatorname{tr}(XY)+\operatorname{tr}(XY^{-1})=\operatorname{tr}(X)\operatorname{tr}(Y). \] These permit propagation of trace preservation from a finite trace set to all words in the generators [1608.08212]. The main innerness theorem states that if \(h:G\to I\) is surjective and trace-preserving, and there exist \(g_1,g_2\in G\) whose associated Möbius transformations have no common fixed point, then there exists \(A\in SL(2,\mathbb C)\) such that \[ h(g)=AgA^{-1}\qquad\text{for all }g\in G. \] When both groups lie in \(SL(2,\mathbb R)\), the conjugating element can be taken in \(SL(2,\mathbb R)\). The proof proceeds by conjugating one generator to diagonal or standard parabolic form, using trace equalities to determine matrix entries of the images, and exploiting the absence of common fixed points to rule out degeneracies [1608.08212]. The same paper develops finite trace parameterizations for finitely generated and finitely presented groups. If \(G=\langle A_1,\dots,A_n\rangle\) and a surjective homomorphism preserves the traces of each generator and of every ordered product \(A_{i_1}\cdots A_{i_k}\) of distinct generators with \(i_1<\cdots